Why is it that handicap games are so scarce on OGS?

I get it now, this makes a lot of sense!

1 Like

Elon is that you??

5 Likes

I play a lot of games with handicap=automatic. Despite the facts and sometimes prejudices there are against handicap go, I prefer to lose to a 14 kyu who gets a two stones head start on a 9x9 instead of a no handicap game which is not really challenging.
So if @flat_chocolate_mount wants to play a (teaching) game with handicap, I am ready.

2 Likes

Thanks!

He’s not wrong, though (I don’t like him as a person, and I deeply disagree with his way of doing things, but in this I think he’s right).

(emphasis mine ^)

I believe I might have a good conjecture to answer that, which is two-fold:

  1. This is part of how statistics works: if someone’s rank oscillates a lot, they should find harder handicap games when their rank is high and easier handicap games when their rank is low; but it might just be the case that on average the games they win in excess cancel out with the games they lose in excess, leading to an overall 50% winrate. This does seem to imply that the rank oscillation somehow succeeds in being unbiased in this sense, i.e. it doesn’t systematically over-rank or under-rank players, though statistics is complicated so who knows.

  2. It might also/alternatively be that the rank fluctuations manage to capture the players’ ability even more accurately and precisely than we think, down to “this player is having a good day so their rank rises” etc. – and if you think about it, if a player is having a good day it also stands to reason that they have a better chance of winning a harder handicap game than usual. So players that play a lot and see their rank fluctuate a lot might actually be the ones for which the rank is most accurate, down to their daily mood.

(This just made me think that perhaps a more “ideal” system would keep the rating as is and use it for matching and automatic handicaps, but the displayed rank [the one next to someone’s username] should be a more stable and averaged out version; but I guess that’s a bit off-topic – also it might be a bit of a PR nightmare for the devs)

3 Likes

Actually this is pretty close to how it currently works ^^

Rating is used for everything - pairings, calculating handicaps/macmahon points etc - while the ranks are simply derived from that said rating because human go players are just more used to using kyu/dan ranks.

1 Like

“It might also/alternatively be that the rank fluctuations manage to capture the players’ ability even more accurately and precisely than we think, down to “this player is having a good day so their rank rises” etc”
This is, I think, a key difference between online play and IRL. Many of us will play online regardless of our current mental condition, whereas for a live tournament or club meeting, we will prep, we will try to be rested, we will take the game more seriously.
Thus one’s online performance is likely to vary greatly from one’s rating or ranking, and the effectiveness of handicapping will suffer as a result. Cest la (online) vie.

5 Likes

Actually, to be clear, my whole thing was prompted by @snakesss linking to this topic, which provides stats showing how the handicap system succeeded in keeping winrates around 50% back in 2020 – then just now I found this one about the 2021 ranking system, which keeps handicap winrates loosely at around 43% for Black* (though I guess that doesn’t necessarily mean the stats remained equivalent, and it might be worthwhile updating them if anyone knows how to and can be bothered).

So the point in discussion was that one would think the handicap system would be defective because of all these factors affecting online play, but surprisingly it didn’t appear to be all that bad (well, at least the 2020 system appeared to be good). And that’s where I came in with conjectures as to why that might be the case :slight_smile:

*by the way, I think the lowering to 43% winrate

might at least partially be caused by the rounding down that happens in the automatic handicap calculation. Since the number of stones is always calculated rounding down to a whole number, more often than not Black will have less stones than the actual rank difference. Down to 43% from 50% is still kind of a lot, though, so there are probably other reasons. In fact, looking more closely at those 2021 updated stats, I’m not sure I would say the handicap system works all that well anymore.

5 Likes

Thank you, that’s the one I wanted to link, did not realize I was linking the old one.

To humor the tangent:

I would actually suggest the lowering of winrate is potentially caused by the rating system in itself,
largely that our analysis of the games prior to implementation was trained on a dataset where matchmaking was made with an older system, which as someone better than me with Machine Learning once pointed out is still enough to create some level of skewing of results, as the matchmaking by design works to create the most evenly matched games it can based only on rating results.

Of course, while this game data should still have been useful in terms of estimating “true ratings” and expected winrates, it also could easily have been “overfit” to the training data from the old system and not perfectly reflected post-implementation

note: this is on top of questions already raised in the past about the measure of fit, which has been argued is statistically questionable

1 Like

Why is it that handicap games are so scarce on OGS?

  • because of the enormous amount of sandbagging on OGS
  • because handicap go is a different game
  • another reason

0 voters

yeah, I was thinking that too, but it does seem to be going more and more off-topic :sweat_smile: also I don’t understand statistics or know enough about this specific issue to be able to discuss it properly.

...buuut, since we're talking about it

I keep thinking along the lines of what I said before:

It looks to me that the 2021 system attempted to solve at least two problems at once: improve rank stability, and get the best possible matchmaking. And it looks to me that what actually ended up happening is that rank stability was improved at the cost of a worse handicap system.

So I keep thinking that we should just separate ranking and rating, fine-tune the rating to make the best possible matchmaking and handicap system, accepting that that entails heavy rating fluctuations, and use a very smoothed out version for the ranking, with the understanding that ranking is just an arbitrary convention that indicates “how strong” you are “as of late”.

A lot of people would be unhappy and confused with such a system, mostly because they’re attached to the idea that the matchmaking and the handicap system should correspond to the ranking, so I don’t expect anybody to listen to me, but… I just really like the idea :pensive:

1 Like

Your choices are not mutually exclusive, which makes it impossible for me to answer the poll. I would say it is a combination of reasons. I previously enumerated two, but did not mention the occurrence of botting, which is similar in effect to sandbagging.

1 Like

Well, I did it on purpose. I realised that there must be more to it and knew it would cause some raised eyebrows. So I specially put another reason as an option, but was surprised that most people point to another reason as cause for the scarcity of handicap games on OGS.
If or when it becomes clear what more reasons there are, I will provide the forums with a more elaborate poll.
:grin:

handicap ruins fuseki
it makes sense only if there is too big rank difference between players, so they at least would have fun with midgame/endgame mechanics of Go.

1 Like

Well, maybe that’s why I don’t dislike handicap. I have exactly zero fuseki knowledge that could be ruined.

3 Likes

I think part of the reason handicap games are uncommon is because they arent encouraged by the defaults on OGS. Automatch defualts to a max rank difference of +/- 3, which will result in low handicap, and custom games default to No handicap.

7 Likes

I don’t think handicap arrangements on 19x19 ruin fuseki. If anything, the player wielding White in a handi game can sharpen his/her local and global fighting skills. I like playing handicap games as White for this very reason. If giving handi stones is unfair, why not just spend the first 50-100 moves playing out a predetermined whole-board sequence that will result in about equal advantage for both sides, then play freely for the rest of the game?

Many say handicap is a different game. Something you need to study especially.

I don’t think so. It’s not a different game, not something like more complex in its specificity.

At reverse, handicap simplifies the game by giving you a way to win. It will give the weaker player the opportunity to win if you follow just basic understanding of using influence. It will teach you to not be a coward but still reasonably cautious.
Keep white in your boxes. Make your stones work together, let him live small while you get big.

That’s very different as an even game where you have less obvious constraints, more choices and where it’s easy to forget those fundamentals.

As long as you don’t commit yourself against the way handicap works, you’ll be able gradually to ameliorate your moves and climb the rating. It’s an extraordinary tool to progress and at the same time will offer some interest to the stronger.

7 Likes